Temperature is a parameter in LLM decoding algorithms that controls randomness in text generation.
Effects of Temperature on Text Generation:
Higher Temperature (>1.0):
Flattens the probability distribution, making lower-probability words more likely.
Increases randomness, resulting in more creative and diverse outputs.
Lower Temperature (<1.0):
Sharpening effect, making high-probability words more dominant.
Produces more predictable and deterministic responses.
Why Other Options Are Incorrect:
(B) is incorrect because temperature does not remove the impact of likely words; it reduces or increases randomness.
(C) is incorrect because temperature affects probability, not speed.
(D) is incorrect because decreasing the temperature narrows the distribution, making text more deterministic.
🔹 Oracle Generative AI Reference:
Oracle AI models allow dynamic temperature control to balance coherence and creativity in text generation.